Un i ve r s i t y o f Kons t an z Depa r tmen t o f E c onom i c s

Credit Spread Interdependencies

of European States and Banks during the Financial Crisis

Adrian Alter and Yves Stephan Schüler

http://www.wiwi.uni-konstanz.de/workingpaperseries

Working Paper Series 2011-24

Credit Spread Interdependencies of European States and

Banks during the Financial Crisis

Adrian Alter , Yves Stephan Schüler∗

This Draft: 22th of June, 2011†

Abstract

This study analyzes the relationship between the default risk of several European states andfinancial institutions during the period June 2007 - May 2010. It investigates how this linkagewas effected by government bailout schemes. We consider sovereign credit default swap (CDS)spreads from seven EU member states (France, Germany, Italy, Ireland, Netherlands, Portugal,and Spain) together with a selection of bank CDS series from these states. Long-run and short-run dependencies between states and their domestic banks are studied within a vector errorcorrection framework and in addition we consider generalized impulse responses. Our mainfindings suggest that in the period preceding government rescue schemes the contagion frombank credit spreads disperses into the sovereign CDS market. After government interventions,due to changes in the composition of both banks’ and sovereign balance sheets, our resultsunderline the augmented importance of the government CDS spreads in the price discoverymechanism of banks’ CDS series. Furthermore, a financial sector shock affects more stronglythe sovereign CDS spreads in the short-run however the impact becomes insignificant at a longhorizon. The variability in linkages between bank and sovereign CDS spreads can be associatedwith differences in state support measures accessed by each bank. Our cross-country analysisreveals noticeable differences in the outcomes of government bailouts.

JEL-Classification: C58, G01, G18, G21.Keywords: credit default swaps, private-to-public risk transfer, bailout programs, cointegration,generalized impulse responses.

∗University of Konstanz, Germany; Department of Economics; Chair of International Finance, Chair of Statistics and Econometrics;adrian.alter@uni-konstanz.de, yves.stephan.schueler@uni-konstanz.de†We are very thankful for comments from Daniel Andrei, Ralf Brüggemann, Günter Franke, Ser-Huang Poon, Peter Raupach and

the participants of the Computational and Financial Econometrics Conference (London, Dec 2010), and the Marie Curie MidtermConference (Berlin, May 2011). Research support (in form of a Ph.D. Fellowship in Risk Management) for Adrian Alter from Eu-ropean Community’s Seventh Framework Programme FP7-PEOPLE-ITN-2008 (grant agreement PITN-GA-2009-237984) is gratefullyacknowledged.

1

“The scope and magnitude of the bank rescue packages also meant that significant risks hadbeen transferred onto government balance sheets. This was particularly apparent in the marketfor CDS referencing sovereigns involved either in large individual bank rescues or in broad-basedsupport packages for the financial sector.”

BIS (December 2008)

1 Introduction

During the recent financial crisis extraordinary measures were taken by central banks and gov-ernments to prevent a potential collapse of the financial sector that threatened the entire economy.However, the effects on the interdependence of the financial and the sovereign sector were widelyunknown. Gray (2009, p. 128) argues that “regulators, governments, and central banks have notfocused enough on the interconnectedness between financial sector risk exposures and sovereignrisk exposures and their potential interactions and spillovers to other sectors in the economy orinternationally”. The lack of theoretical macroeconomic models that are able to incorporate con-tagion mechanisms between government and financial sector amplified the uncertainty related toimplications of government interventions. Nevertheless, regulators and policy makers need to un-derstand the complex dynamics of the risk transmission in order to be able to formulate effectivepolicies and to be aware of the risk transferred from the financial to the public sector.

Why should banks’ and sovereign default risk be related? As pointed out by Gray et al. (2008), us-ing arguments from the contingent claims analysis (CCA)1, there are several channels linking thebanking and sovereign sector, which are impacted on by the implicit as well as the explicit guar-antees. On the one hand, a systemic banking crisis can induce a contraction of the entire economy,which weakens public finances (e.g. by a decrease of the present value of taxes) and, thus, trans-fers the distress to the government. This contagion effect is amplified when state guarantees existfor the financial sector. As a feedback effect, risk is further transmitted to holders of sovereigndebt. On the other hand, macroeconomic imbalances in a specific country lead to rising sovereignspreads and a devaluation of the government debt that is mirrored in banks’ balance sheets.

Empirically, this interconnectedness has been detected in the context of the recent financialcrisis as well. For instance Gerlach et al. (2010) find that as a consequence of macroeconomic im-balances, especially for peripheral European countries (e.g. Greece, Ireland), a jump in bond andcredit default swap (CDS) spreads for sovereigns is further transmitted to the banking sector. Fur-thermore, the authors claim in general that systemic and sovereign risk became more interwovenafter governments issued guarantees for banks’ liabilities. Government interventions proved torestore confidence, as the explicit bailout programs transferred partially credit risk to the publicsector. Moreover, Dieckmann and Plank (2010) present evidence for a private-to-public risk trans-fer in the countries where governments interfered and stabilized the financial system after the

1This approach is based on Merton’s and Black-Scholes’(1973) option pricing work. It can also be employed formeasuring the sovereign-banks interaction, taking into account the implicit and explicit contingent liability for thefinancial system.

2

Lehman Brothers’ event. Banks and sovereign CDS became closer linked through their balancesheets: banks holding important amounts of government debt and states bearing vital contingentliabilities for the financial system.

This paper studies the lead-lag relation between government’s and bank’s default risk with afocus on the effect of government rescue packages in the midst of the last financial crisis. Basedon our empirical analysis we shed light on the implications of the private-to-public risk transfermechanism. First of all, we research whether prior to government interventions an increase indefault risk of banks and states originates mainly from the financial sector. Furthermore, we tryto answer if public contingent liabilities to the financial sector impacted on government’s defaultrisk. In tandem, this study examines whether default risk of the banking sector is influenced bythe sovereign default risk. Finally, our paper investigates the following two questions: 1) Does thedegree of a bank’s participation in a national rescue scheme influence its dependency on the de-velopment of the sovereign spread? 2) Are country specific characteristics reflected in the impactof government bailout programs?

Our main results emphasize the systemic implications of the financial crisis. In the period preced-ing government interventions the contagion from bank credit spreads disperses into the sovereignCDS market. This finding can be interpreted as evidence for the systemic feature of the recentfinancial crisis. The implied risk of default spills over from the financial system to the entire econ-omy and questions the government capacity to repay its own liabilities. Another conjecture isthat market participants (especially government bond holders) saw an implicit guarantee for thefinancial sector before the extreme event of Lehman Brothers’ default. After government interven-tions, due to changes in the composition of both banks’ and sovereign balance sheets, our find-ings underline the augmented importance of the government CDS spreads in the price discoverymechanism of banks’ CDS series. Furthermore, a financial sector shock affects more strongly thesovereign CDS spreads in the short-run however the impact becomes insignificant at a long hori-zon. We suggest that the variability in linkages between bank and sovereign CDS spreads canbe associated with specific state support measures accessed by a bank. Lastly, our cross-countryanalysis reveals noticeable differences in the outcomes of government interventions.

Methodologically, we investigate the short-run and long-run relationships between govern-ment and banks’ CDS, as they provide a proxy for the risk of default.2 We conduct this analysis byapplying the theory of cointegration, Granger-causality, and impulse responses to CDS series, bywhich we argue to capture changes in the dynamic relation between government and bank creditrisk. We consider sovereign CDS from seven EU member states (France, Germany, Italy, Ireland,Netherlands, Portugal, and Spain) together with a selection of bank CDS from these states. Wedivide the analyzed period, i.e. June 2007 until May 2010, into two periods: 1) before state bailoutprograms and 2) during and after government interventions.

As also argued by Markose et al. (2010), there are only a few empirical studies that deal indepth with linkages within the financial system, especially through the usage of CDS spreads,

2The objective of this paper is not to investigate the accuracy of this proxy. Our research design takes this link asgiven, even though there might have been distortions in this proxy during the last turmoil.

3

although credit derivatives played an important role in the recent turmoil. Briefly, a CDS is abilateral agreement that has the role to transfer the credit risk from the “protection buyer” tothe “protection seller”. The former party pays a periodic fee to the latter party (the credit-risktaker), and in return is compensated in case of default of the underlying entity (or similar creditevent), with a payoff3. The reference entity is not a party to the credit default swap and can bea corporation, a sovereign, an index, or a basket of assets that bears credit risk. The CDS spreadrepresents the insurance premium and is paid quarterly until either the contract ends or at thearrival of a credit event. CDS markets were intensively used as a proxy for credit risk in the lastdecade. In normal times, CDS spreads, compared to bond prices, represent a straight forwardand efficient measure for default risk that incorporate other risks (e.g market risk) as well. CDSvolumes, in terms of total notional amount, peaked at $57 trillion in June 2008.4

The structure of this paper is organized as follows. In section 2 we discuss studies are relatedto our research. Section 3 addresses our hypotheses, the data, our sub-sample selection procedure,and the econometric methodology. In Section 4 we present our results and Section 5 concludes.

2 Related Literature

As already indicated, at this moment there are very few papers that actually analyze the effectof the financial crisis and the emerging government policies on the risk of default of financial orsovereign entities. One of these studies is undertaken by Schweikhard and Tsesmelidakis (2009).Relying on a structural model5 they conclude that credit and equity markets decoupled duringthe financial turmoil. They find support for the “too-big-to-fail” hypothesis, as some companies’debt holders benefited from government interventions and a shift of wealth from taxpayers to thecreditors took place during the bailout programs. During the crisis some other factors might haveinfluenced CDS prices (e.g. counterparty or liquidity risk). Researching on the determinants ofCDS changes Collin-Dufresne et al. (2001) find that credit spreads are mostly driven by a system-atic factor; however they are not able to identify it, although they consider as candidates severalmacroeconomic variables and firm specific factors. Berndt and Obreja (2010) study the factors forEuropean corporate CDS and identify the common factor, that explains around 50% of the vari-ation, as the super-senior tranche of the iTraxx Europe index. They name it “economic catastropherisk”, which can be interpreted as a total disfunction of the economy or as a sequence of sovereigndefaults.

Similar to our study, Dieckmann and Plank (2010) find evidence for a private-to-public risk

3In the case of cash settlement only the difference between the par value of the bond (notional amount of the loan)and its recovery value when the credit event occurs is paid in cash by the protection seller. In the case of physicalsettlement the par value is paid in exchange for the physical underlying bond.

4According to the Bank for International Settlements (BIS, http://www.bis.org/statistics/otcder/dt1920a.pdf) at the end of June 2010 this market stood at around $30 trillion in terms of notional amounts out-standing. During this period the gross market value halved from about $3.2 trillion to $1.6 trillion.

5We use the term structural model to indicate a framework for pricing CDS spreads. In more detail, researcherstry to replicate market prices by employing structural models or reduced-form ones. In this context, the well knownMerton model can be regarded as a benchmark.

4

http://www.bis.org/statistics/otcder/dt1920a.pdfhttp://www.bis.org/statistics/otcder/dt1920a.pdf

transfer for the countries with government interventions in the financial system. By employingpanel regressions the authors analyze the determinants of changes in sovereign CDS spreads andfind that both domestic and international financial systems bear an important role in explainingthe dynamics of the CDS spreads. They also argue that countries in the EMU6 are more sensitiveto the health of the financial system than non-EMU countries. Ejsing and Lemke (2009) use aseemingly unrelated regressions framework with time-varying coefficients in order to investigatethe co-movement between the CDS spreads of Euro area countries and banks. Before October 2008the common risk factor that explains much of the variation of these spreads is the iTraxx CDS indexof non-financial corporations. The authors find that government bailout and guarantee programsfor the financial sector induced a drop in the credit spreads for banks but a jump in governments’CDS spreads. Both papers identify as a structural break mid-October 2008, when first massivebank rescue packages were set up after the Lehman Brothers’ default. Our study extends theiranalyses by providing details on the risk-transfer mechanism in several European countries viajoining a cointegration and a generalized impulse responses framework. Our research is one ofthe first studies that is able to emphasize the important changes in these linkages. We argue thatour findings are mainly a consequence of the bailout programs and the heterogeneity in effectsacross European states is a result of the relative differences between the rescue schemes.

Using the contingent claims analysis for financial institutions, Castren and Kavonius (2009)identify the main channels that propagate risk within the euro area, by constructing a risk-basednetwork from bilateral sector-level exposures. They claim that the financial sector plays an impor-tant role in transmitting shocks to the other sectors. From a different perspective, the latter paperis able to reveal the mechanism that interconnects governments and financial institutions throughthe intermediary of credit risk channel.

Furthermore, there are studies that solely investigate the sovereign bond market. Using aGARCH-in-mean model, Dötz and Fischer (2010) analyze the EMU sovereign bond spreads andfind that the implied probability of default reached unprecedented values and the increased ex-pected loss component made some sovereign bonds to loose their status of “safe haven” invest-ment. As a structural break they use the rescue of Bear Stearns in March 2008, as after this momentthe market perceptions of sovereign risk changed dramatically. Gerlach et al. (2010) analyze thedeterminants of Euro area sovereign bond spreads. They show that the size of the banking sectorhas an important explanatory value for the changes in bond spreads, suggesting that markets per-ceive countries with an important stake of this sector at higher risk of stepping up and rescuingthe banks in the system and thus increasing the debt burden. Employing a dynamic panel Atti-nasi et al. (2009) highlight the main factors that explain the widened sovereign bond spreads insome Euro area countries for the period that covers the core part of the financial crisis in Europe.They find that sovereign bond yields are explained by expected fiscal fundamental indicators, aliquidity measure, and a global risk aversion measure.

The present study is also related to the literature that analyzes contributions to the price dis-

6European Monetary Union

5

covery of bond and CDS markets. For example Dötz (2007) studies European corporate entitieslisted in the iTraxx Europe index in a time-varying setup and finds support for a relatively equalcontribution to the price discovery of both markets. During the credit turmoil in 2005, CDS mar-kets’ contribution fell dramatically, leaving space for the interpretation that CDS and bond mar-kets are decoupling during turbulent times and that CDS prices might lag behind bond marketsin price discovery for corporations. Since then, CDS markets consistently developed in terms ofvolume and market participants. Focusing on Euro area sovereigns, Fontana and Scheiche (2010)identify the main determinants of the bond and CDS spreads. They include in the set of explana-tory factors proxies for market liquidity and global risk appetite and these are found as significant.Furthermore, they employ a lead-lag analysis for bond and CDS markets and find that for westerneconomies7 the cash market dominates, while for peripheral countries8 the CDS market is moreimportant in terms of price discovery.

3 Hypotheses, Data, and Econometric Methodology

3.1 Hypotheses

In this subsection we develop the hypotheses which are considered in the empirical part of thisstudy. From a theoretical point of view, we firstly describe the main transmission channels thatemerge when either a (systemic) banking crisis develops or sovereign distress appears. Based onGray (2009) and IMF (2010), we present both directions of the contagion mechanism.

If a financial institution faces funding and/or liquidity issues, this can trigger a sharp risein its risk of default and further may have some specific contagion effects: (I) the bank cannotpay its obligations to another financial counterparty which in turn can set off funding/liquiditydifficulties for the latter and an increase in its perceived default risk; (II) the state might intervenein order to prevent bankruptcy of banks via a private-to-public risk transfer. The latter translatesinto a higher probability of default for the state and a decrease in the default risk of the financialinstitution. If (I) occurs, difficulties within the entire financial system (e.g. systemic banking crisis)might arise and translate into a contraction of the economy, which weakens public finances (e.g.by a decrease of the present value of taxes) and, thus, the sovereign risk of default increases.Furthermore, when the government balance sheet includes financial guarantees (i.e. (II)) thesecontagion effects are amplified (e.g. contingent liabilities increase).

In the case of a country’s distress, in the first wave, the contagion to other entities can be trig-gered via three direct channels (cf. Chapter 1. of IMF (2010)): (i) from the affected state to othercountries that are highly interconnected through bilateral trade, or share similar problems (e.g.public deficit, funding needs, etc.); (ii) from the distressed country to domestic banks as the mar-ket value of government bonds held by these banks decreases, their funding cost jumps, a possiblefollow-up currency depreciation influences exposures in foreign currency and government contin-

7France, Germany, Netherlands, Austria, and Belgium.8Greece, Italy, Ireland, Spain, and Portugal.

6

gent liabilities lose credibility; (iii) from the impaired state to foreign banks, that hold importantgovernment (or banks) bonds (or other assets) from the affected country. The second wave ofspillovers is via indirect linkages (rising counterparty risk, increasing collateral demands or regu-latory capital, eroded market confidence, etc.). In our study channels (i) and (ii) are considered.Before government interventions, we argue that financial sector issues had a systemic component,which in turn led to a deterioration of the selected economies, i.e. contagion mechanism (I). Thus,the rising risk of default for banks (impairment of the balance sheet) had a similar indirect effecton the development of the risk of default for governments. Additionally, state interventions inresponse to financial sector problems were possibly expected by market participants, which inturn could have led to a prior adjustment of the perception of sovereign default risk. Anotherpotential driving force in this adjustment process is changes in investors’ risk aversion. Therefore,in this period sovereign credit risk should be sensitive to movements in the default risk of banks,while the risk originating from the governments is irrelevant. Based on these arguments, our firsthypothesis is formulated as follows:

Hypothesis 1. Prior to state interventions changes in the default risk of banks impact on the defaultrisk of European governments but not vice-versa.

After government interventions, states do not only bear an asset exposure to the banking sector(e.g. corporate taxes collected) but their balance sheets contain contingent liabilities (e.g. govern-ment guarantees) as well. Thus, the sensitivity of government default risk to the developments inthe banking sector risk is expected to increase. Furthermore, through the credibility of governmentcontingent liabilities, changes in government default risk directly impact on the perceived risk offinancial institutions.

Hypothesis 2 (a). After the private-to-public risk transfer, changes in the default risk of banks impacton the default risk of states stronger than before bailout programs.

Hypothesis 2 (b). Subsequent to bailout programs, an increase/decrease in government’s default riskaffects the default risk of the domestic banks in the same direction.

We infer from the above portrayed mechanism that the default risk of governments and banksare directly linked. Furthermore, if a government intervenes by offering guarantees to the finan-cial sector, it is vital for the state to be able to credibly provide funds for preventing a systemiccrisis. As public finances are already weakened by the contraction of the economy, which arises asa consequence of the problems in the banking sector, the health of the economy seems crucial tolimit the threat by the vicious circle described above. Thus, we argue that rescue scheme character-istics9, and the degree of a bank’s participation in public aid are important factors for explainingdifferences in impacts of bailout programs across countries and banks.

Hypothesis 3. The sensitivity of the bank to the government risk of default increases with the perceivedrisk transferred from the bank to the government.

9A detailed discussion on bailout specific characteristics is provided in the next subsection.

7

Hypothesis 4. Government rescue schemes across European countries influenced heterogeneously theinterdependence between states’ and banks’ default risk.

3.2 Bailout Specific Characteristics

In order to draw conclusions on the effects of the bailout schemes across the selected countries, werelate our analysis to the specific program provided in each country. Hence, we look at the magni-tude of different measures utilized by each country, while additionally considering the particularaid for each bank. Following Stolz and Wedow (2010), we categorize the general set of measuresemphasizing the differences and similarities across countries. Even though there is a discrepancyin the number and types of institutions involved in the banking crisis management, there is lessvariation across countries in what types of support measures were applied. The financial aid pro-grams can be classified into four broad categories: capital injections, guarantees for bank liabilities,asset support programs, and deposit insurance. From another perspective support programs canbe divided into ad hoc measures and national schemes. Ad hoc measures (outside schemes) rep-resent individually tailored rescue programs for specific banks, whereas national schemes denotesystem-wide programs. The main advantages of the latter are high transparency and accessibilityamong financial institutions. Table 1 refers to the aggregated commitment within the seven coun-tries studied in this paper. Details on banks’ specific participation in the public aid schemes aredescribed in Section 4.

Table 1: Government Support Measures to Financial Institutions (October 2008 - May 2010)

Country Capital injection Liability guarantees Asset support Total DepositGuaranteed Other commitment insurance

Within Outside issuance of guarantees, Within Outside as %Schemes Schemes bonds loans Schemes Schemes 2008 GDP in EUR

France 8.3 (21) 3 134.2 (320) 0 - (-) - 18% 70,000Germany 29.4 (40) 24.8 110.8 (400) 75 17 (40) 39.3 25% UnlimitedIreland 12.3 (10) 7 72.5 (485) 0 8 (90) - 319% UnlimitedItaly 4.1 (12) - - (-) 0 - (50) - 4% 103,291Netherlands 10.2 (20) 16.8 54.2 (200) 50 - (-) 21.4 52% 100,000Portugal - (4) - 5.4 (16) 0 - (-) - 12% 100,000Spain 11 (99) 1.3 56.4 (100) 9 19.3 (50) 2.5 24% 100,000

Note: Numbers in brackets denote totally committed funds. All amounts are in billions of EUR, except for the last two columns.Within schemes refer to a collective bailout program that can be accessed by any bank that fulfills the requirements for that particularaid scheme. Outside schemes refer to individually tailored aid measures (ad-hoc schemes). Source: Stolz and Wedow (2010)

Based on the ratios of total commitment to GDP, our selected countries can be ranked (fromhigh to low): Ireland, Netherlands, Germany, Spain, France, Portugal, and Italy. Furthermore,the set of countries can be clustered into three groups: Ireland (high commitment); Netherlands,Germany, Spain, France (medium commitment); Portugal and Italy (low commitment).

8

3.3 Data and Sub-Sample Selection

We use daily CDS spreads collected from Datastream10, for seven European countries togetherwith two banks from each country, i.e. in total 21 institutions: France (FR), BNP Paribas (BNP),Société Générale (SG), Germany (DE), Commerzbank (COM), Deutsche Bank (DB), Italy (IT),Intesa Sanpaolo (ISP), Unicredito (UCR), Ireland (IR), Allied Irish Banks (AIB), Bank of Ireland(BOI), Netherlands (NL), ABN Amro Bank (ABN), ING Group (ING), Portugal (PT), Banco Com-ercial Portugues (BCP), Banco Espirito Santo (BES), and Spain (SP), Banco Santander (BS), BancoBilbao Vizcaya Argentaria (BBVA).11 All banks are important financial institutions, with most ofthem belonging to the iTraxx Europe index (8 out of 14). The selection of banks and sovereignsCDS was restricted by data availability. In terms of CDS spreads we decided to use contracts onsenior unsecured debt with 5 years maturity, as they are known to be the most liquid ones.

Our sample covers the time span from 1 June 2007 until 31 of May 2010 and includes 772observations of daily data for each of the selected series.12 Graphs of the development of the CDSlevels for all 21 series are presented in Appendix B (in subsections for each country). Prior to theeconometric analysis, we log-transform the CDS levels as suggested by Forte and Pena (2009). Wefurther motivate this step by relatively low levels of the CDS for the sovereigns in the first stagesas compared to the last stages (wide data range) and the positive skewness.

Our aim is to analyze the linkages between bank’s and sovereign CDS spreads in a two sub-period setup: 1.)before government interventions and 2.)during and after government aid schemes.In order to capture other structural breaks, we follow BIS (2009) and divide the entire time spaninto six stages13. We group the first two stages (i.e. Stage 1+2) to form the sub-period beforegovernment interventions and the last three stages (i.e. Stage 4+5+6) to constitute the sub-periodduring and after the government aid schemes. Stage 3 is regarded as a period of market adjust-ments which we neglect. When issues concerning structural breaks appear in our stability analysis(see 3.4), we analyze stages individually or in two combinations (i.e Stage 4+5, Stage 5+6).

The first stage runs from June 2007 until mid-March 2008 and contains 203 observations. Thisperiod is characterized by financial stress which has been triggered by fears on losses due to USsubprime mortgage loans and the spillover to European banks (e.g. IKB Deutsche Industriebank,BNP Paribas). As a result, the financial system is weakened and finally by the liquidity shortage ofBear Stearns the second stage emerges. This time span consists of 126 observations and is definedto evolve until mid-September 2008, when the problem of funding opens the possibility for insol-vency of banks, which leads to the decline of Lehman Brothers and to the third stage. BIS (2009)defines this short stage to range from mid-September until late October of the same year. As thisshort stage includes only 30 observations we do not focus on it in our analysis. When Lehman

10We downloaded CDS data from Datastream, which in turn is provided by Credit Market Analysis (CMA).11In parentheses are defined the abbreviations for later use.12In the case of Ireland the sample starts on 10/04/2007 because of inconsistencies with the data obtained from

Datastream.13BIS (2009) covers only our first five stages, starting with 1 June 2007 until 31 March 2009 when Stage 5 emerges. For

the time span that was not included in the latter study we define a sixth stage. The last stage is selected to start basedon developments in the sovereign CDS market at the end of 2009.

9

Brothers files for bankruptcy protection a global crisis is triggered, in which more and more fi-nancial institutions are threatened by risk of default. During this time already first governmentpolicy measures are taken14. Additionally, coordinated actions by major central banks try to takecontrol over the situation. The fourth stage is defined to be from late October 2008 to mid-March2009 and contains 98 observations. This period is marked by concerns about a deepening of theglobal recession. At this point in time, the European Commission, by issuing guidelines15 for Eu-ropean states, gives the green light for government bailout programs. Stage 5 starts in mid-March2009 when the first signs of recovery appear. Announcements of central banks concerning balancesheet expansions, the range and the amount of assets willing to be purchased lead to an importantfinancial markets relief. The end point of the fifth stage is the 30th of November 2009, right beforethe inception of sovereign debt issues in Europe. This stage includes 143 observations. The laststage of our sample (i.e. Stage 6) lasts from the end of 2009 until the end of May 2010 and consistsof 172 observations. This period is labeled by concerns related to European sovereign debt. InMay 2010 European governments set up a rescue fund for aiding states in trouble, that temporaryreleases the pressure on some problematic countries.

3.4 Econometric Methodology

In order to analyze the dynamics of the short- and long-run interdependency between selectedCDS series, this study employs a vector error correction (VEC) and vector autoregressive (VAR)framework. Within the first class of models, variables can be analyzed that share a commonstochastic trend. If two or more series share at least one common stochastic trend they are calledcointegrated (cf. Engle and Granger (1987)). Besides interpreting the cointegration relations, weadditionally conduct tests on Granger-causality and consider impulse responses in order to de-scribe the entire dynamics between the CDS spreads.

During tranquil times we believe that the CDS series of the financial as well as governmentsector are stationary. However, during times of market turmoil we argue that both CDS series (i.e.bank and sovereign CDS) are impacted by the same stochastic trend, because both are linked bychannels as expressed in Subsection 3.1.

Taking into account that integration of at least order one of the series is a necessary conditionfor a common stochastic trend, we firstly apply the standard unit root (stationarity) testing pro-cedures to the respective time series in each sub-sample. The tests utilized are the Augmented-Dickey-Fuller (ADF), Phillips-Perron (PP), and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test.All of the latter include an intercept because we disregard the possibility of a zero mean or trendstationary process. The latter process is not considered as it is economically unreasonable to as-sume that CDS series rise perpetually.

We do not analyze systems of CDS series in a vector error correction model (VECM) if there is

14E.g. UK authorities intervene in an attempt to relief the pressure on financial stocks through the suspension of shortselling on some financial products.

15cf. IP/08/1495

10

evidence that one or more series are stationary as in this case they cannot share a joint stochastictrend. For detecting a common stochastic trend, this study considers on the one hand Engle-Granger ADF test and on the other hand Johansen’s trace and maximum eigenvalue tests16. Thetests used in this study focus only on the setup with a restricted constant, as argued before, anydeterministic trend in the variables or cointegration relation is economically unjustified. Whencommon stochastic trend(s) are detected by one of the previous tests, we model the series in aVECM framework. In case the tests do not reveal a cointegration relation, we divide the periodinto sub-periods, because we attribute this result to structural breaks.

For determining the optimal lag order p, we minimize one of the common information criteria:Aikaike information criterion (AIC), Hannan Quinn criterion (HQ), Schwarz criterion (SC), andFinal prediction error (FPE). As a goodness-of-fit, we check for serial correlation of the residualsand set the optimal lag as a result of the two procedures.17 A VECM with p− 1 lags can be writtenas follows:18

∆yt = Π0y0t−1 + Γ1∆yt−1 + · · ·+ Γp−1∆yt−p+1 + ut = αβ0′y0t−1 +p−1∑i=1

Γi∆yt−i + ut, (1)

where

y0t−1 ≡

[yt−1

1

]and Π0 ≡

[Π−Πµ0

],

where yt is a column vector ofK variables integrated of order one, ∆yt represents the vector ofvariables in first differences, Π and Γi are coefficient matrices, µ0 is a constant, and ut is assumedto be wn(0,Σu)19. The Π matrix captures the long-run and the Γi (i = 1, . . . , p − 1) matrices theshort-term dynamics. If we find r cointegration relations, it means that Π (a K × K matrix) hasa reduced rank (r < K) and can be decomposed such that Π = αβ′

(K×r)(r×K). The elements of α

are known as adjustment or loading coefficients, while β is known as the contegration vector, meaningthat β′yt is stationary and forms the system of cointegration equations. We estimate the model byJohansen’s maximum likelihood procedure.

From (1), a bivariate VEC system with a sovereign CDS spread (in short ’Sov’) and a selecteddomestic bank CDS spread (in short ’Bk’) is given by the following representation:(

∆cdsSov,t∆cdsBk,t

)=

(αSov

αBk

)(βSovcdsSov,t−1 + βBkcdsBk,t−1 + β0) + · · · , (2)

16Only Johansen’s tests serve as an indication of the cointegration rank.17We also look at the plots of cointegration relations in order to check whether these can be argued to be stable. The

plot is expected to a show a time series that fluctuates nicely around some mean18We use the notion of p− 1 lags, to remind of the fact that a VECM(p− 1) has a VAR(p) representation.19wn stands for “white noise” and refers to a discrete time stochastic process of serially uncorrelated random variables

with the above mentioned first two moments.

11

where cds = logCDS. As it is common, βSov is normalized (i.e. βSov = 1) and only βBkis estimated. β-coefficients describe the long-run relationship between banks and sovereign log-CDS spreads. The loading coefficients α measure the speed of adjustment with which a particularCDS adjusts to the long-run relationship. The adjustment forces start acting, whenever the long-run relation (defined by β′yt−1 = 0) is out of equilibrium, i.e. if β′yt−1 6= 0. In case that αSovis significant and has an opposite sign to βSov (i.e. in our setup αSov < 0) it means that the“Sovereign” is driven by the error correction mechanism. Or differently, that it adjusts back tothe long-run equilibrium defined by β′yt−1 = 0, whenever β′yt−1 6= 0. Equivalently, when αBkis significant and has an opposite sign to βBk, it shows the speed of adjustment of the “Bank”to the equilibrium. With both α-coefficients significant and opposite signs to their respective β-coefficients, the variables are said to be in a real cointegration relationship; both series are takingpart in the error correction mechanism. Whenever one of the α-coefficients is not significant, itmeans that the respective variable can be argued to provide the stochastic trend that determinesthe long-run relation. This can be formally tested using a Likelihood Ratio test through a zerorestriction on this parameter. If the restriction cannot be rejected, the variable of the respective αcoefficient is called weakly exogenous. Furthermore, it is not adjusting at all in case that the variablesare not in long-run equilibrium, β′yt−1 6= 0. Whenever one α coefficient is significant but with thesame sign as the respective β parameter, the variable is said not to be part of the error correctionmechanism as the forces in the model do not attract both series back to equilibrium. Series in thissetup can only define a long-run relation if the variable that is in a formal error correction relationadjusts faster to the new equilibrium than the other one. One can think of this phenomenon ina way that the variable which is not part of the error correction mechanism moves the entireequilibrium (i.e. when the variable increases in value the long-run equilibrium will be establishedwith both series achieving a higher value). In the literature the term overshooting20 is used todescribe this occurrence.

As already mentioned the time frame of our analysis may harbor structural breaks. Especiallyin the context of the analysis of long-run relations it is vital to test for stability of the cointegrationspace. For this we conduct stability analyses on the cointegration relations found. We apply thetest proposed by Hansen and Johansen (1999) using recursively estimated eigenvalues. Cointe-gration relations, which do not pass this test are excluded from the argument of our paper.

Furthermore, we embed the sub-sample structure through a top-down approach. We considerthe first and the second stage together as the relevant period (e.g. Stage 1+2) before governmentinterventions. Stage 3 is regarded as a period of market adjustments which we neglect. Stagesfour, five, and six together mark the post government intervention period. Within the cointegra-tion analysis we only refer to further sub-samples if the stability test requires. Thus, we firstlyapply our cointegration methodology to the entire period preceding and the entire period aftergovernment intervention. Only if stability of the cointegration space is rejected, lets say for stagesone and two together, we estimate further VECMs using two new samples, i.e. stage one and

20For a discussion of a model with overshooting please refer to Hansen and Johansen (1998).

12

stage two separately. Cointegration results are only reported for the stages that pass the stabil-ity test using the 1% critical value as a decision boundary. Results from Granger-causality andimpulse response analysis are reported for the entire period before and during/after governmentinterventions, as our main goal lies in contrasting the differences in dynamics of exactly these twosettings.

After describing the basics of the analysis of long-run relations, we proceed with short-rundependencies steps. Firstly, we run the test for Granger-causality, which applies a Wald test on lagaugmented VARs as proposed by Dolado and Lütkepohl (1996). This test is chosen as it guaran-tees the validity of the asymptotic distribution of the test statistic even when there is uncertaintyabout the cointegration properties and stationarity of the variables. In more detail, we estimate anunrestricted VAR in levels with p+ 1 lags, i.e.

yt = ν +p+1∑i=1

Aiyt−i + ut,

where ν is a vector of intercepts. Now consider the coefficient matrix Ai, which can be writtenin our bivariate setup as

Ai =

[αSovSov,i αSovBk,i

αBkSov,i αBkBk,i

],

then using the order of variables presented above the bank does not Granger-cause the gov-ernment CDS series if and only if the hypothesis H0 : αSovBk1,i = 0 for i = 0, . . . , p cannot berejected.

Just looking at short-run or long-run dynamics separately might be misleading. For a globalview on the interrelations of the series another tool is required. Thus, we employ generalized im-pulse responses (GIR) as proposed by Pesaran and Shin (1998). Routinely the analysis of impulseresponses is carried out via the application of the Cholesky decomposition. However, the re-searcher has to specify some causal ordering of the variables. In our case, a theory which specifiessuch ordering is hard to justify, especially in the context of daily data. Based on this argument, wedecided to use GIR because no ordering is necessary and contemporaneous relations are allowedfor. One can regard GIR as the effects of a shock in the structural error of the variable that is or-dered first in the system of orthogonalized impulse responses. To model the uncertainty aroundour point estimates of the impulse responses we decided to use Hall’s bootstrapped percentileconfidence interval (95% level) using 2000 replications (cf. Hall (1992)).

For a stationary VAR(p) process the impulse responses are obtained from the vector movingaverage (VMA) representation:

yt = µ+∞∑i=0

Φiut−i,

where µ is the mean vector and Φi the matrix of the VMA coefficients, which can be calculated

13

in a recursive way from the VAR coefficient matrices,Ai. Impulse responses are obtained by takingthe partial derivative with respect to the shock of interest, i.e. ∂yt/∂ut−i = Φi. In other words, φjk,i,which is defined as the jk-th element of Φi, carries the response of the j-th variable (j ∈ (Sov,Bk))of the system to a shock of size one in variable k (k ∈ (Sov,Bk)), i periods ago. This holds ifthe effect of the shock is not influenced by other distortions in the meanwhile. A cointegratedVAR does not have a ususal VMA representation, nonetheless the matrices of interest, Φi, can becomputed as for the above case. In this setup, however, we have that the effects of shocks maynot die out asymptotically (Lütkepohl, 2007, pp. 18-23, 52,263). The generalized impulse responsefunction can be written in our bivariate setup as follows:

(ψSovSov(n)ψBkSov(n)

)= σ−1/2(Sov,Sov)ΦnΣu

(10

),

(ψSovBk (n)ψBkBk(n)

)= σ−1/2(Bk,Bk)ΦnΣu

(01

)

where σ(j,k) is the variance related to the error of variable j, k (again j, k ∈ (Sov,Bk)) andn denotes the period after which the impulse has occurred. ψSovBk (n) denotes for example theresponse of the sovereign logCDS to a shock in Bk n periods ago. The exact interpretation ofthe impulse responses follows the usual reading for semi-elasticities. E.g. taking into accountthat Φ0 = IK , an impulse in variable j in period 0 means a unit increase in the structural errorthat leads to an increase of the respective CDS series by σ1/2(j,j)%. In order to enable more easilya comparison of the results across banks and countries, we standardize each series of impulseresponses, i.e. the responses caused by the same shock are divided by the standard deviation ofthe impulse variable. In the example above, our responses would be divided by σ1/2(j,j), so that theinitial response of the j-th variable to its own shock is equal to 1 or 100% of the initial shock of sizeone standard deviation. Responses can, thus, be interpreted as percentages of the initial shock inthe impulse variable.

4 Results

This section presents the results for long-run and short-run relationships and in addition considersthe generalized impulse responses. Firstly, the cross country analysis is presented and secondlywe report the specific results for three countries. In the appendix, Table 2 shows the results of allcountries for Granger causality tests, Table 3 outlines the results from our cointegration analysisand Table 4 summarizes the generalized impulse responses for all countries.

4.1 Cross-country Analysis

The effects of the shocks originating from the banking sector on the sovereign CDS series are allsignificant in the long-run in the period before government interventions. This fact is argued bythe systemic component of the crisis that originated from financial institutions and spilled overto the sovereign CDS market. As pictured in Figure 1, and referring to the long-run effects (after

14

Figure 1: States’ Responses to a Banking Sector Shock: Stage 1+2 (Before Government Interventions).

0.2

.4.6

.8

0 5 10 15 20Day

FR (BNP) DE (COM) IR (BOI)IT (UCR) NL (ABN) PT (BCP)SP (BS) Average

Note: sources of the banking shock are written in parentheses. The “Average” line represents the mean of the sovereign responsesfrom a shock in the seven bank CDS spreads.

22 days), countries can be separated into two groups: INNER composed by FR, DE, NL (withresponses over 70%, after 22 days) and OUTER that consists of IR, IT, SP, PT (with responses below50%). The solid line (the “Average” impulse responses) separates the two groups.The results forthe INNER group can be argued to be a result of a weak interest in the CDS market for insuringagainst the default of these countries in the period before Lehman Brothers’ collapse, that couldhave led first to mispricing and then followed by a strong repricing effect. On the other hand,in this period, OUTER countries were already at levels closely linked to their domestic banks’CDS spreads. Public imbalances and high debt burdens were already priced for the latter group.This result is also emphasized in Figure 2 (Right), which contains the long-run (at day 22) effects,and in which we find exactly these two clusters (INNER and OUTER) of banks before governmentinterventions (circles). Additionally, this graph reveals that even the responses of the banks to theirown shock can be grouped into these groups. The impulse of the OUTER group has a smaller effecton itself (y-axis). In general comparing shocks from the banking sector before and after the bailoutprograms, we see that the effects on states’ CDS spread are mostly temporary in the period after,i.e. the effect dies out over time (+ in Figure 2 (Right)) as most long-run effects are not significant21.In the long run, shocks from the banking sector has a smaller impact in the period after (cf. Figure2 (Right): circles relative to pluses). Nonetheless, we observe that Portuguese banks and ISPbecome more sensitive to a shock from the banking sector in the long-run (at day 22). Beforebailout programs, states CDS series are affected progressively by a financial sector shock. Irelandseems to be most successful in limiting the risk of default stemming from the banking sector in theshort-run, as both bank shocks have the lowest effects on itself and the sovereign among all banks

21At day 22 a shock from SG has no significant effect both on itself and on the state. BNP, COM, DB, AIB, BOI, UCR,ABN, and BS have no effect on the sovereign CDS, but the response remains significant on itself.

15

(+ BOI, + AIB (lower left corner) Figure 2 (Left)).

Figure 2: Impulse from the Banking Sector: Effects on Bank’s CDS (y-axis) and Sovereign CDS (x-axis): atday 1 (Left) and day 22 (Right).

.8.9

11.

11.

21.

3B

ank

− R

espo

nse

(Im

puls

e V

aria

ble)

0 .2 .4 .6 .8 1State − Response

Before After

SG

BNP

COM

ABN

UCR

ISP

BES

BCP

BBVA

BS

ING

DB

ING

BNPABN

BBVA

BS

AIB

SG

BES

DB

UCR

ISP

BCP

COM

BOI

BOI

AIB

.2.4

.6.8

11.

2B

ank

− R

espo

nse

(Im

puls

e V

aria

ble)

−.5 0 .5 1State − Response

Before After

ABN

ING

DB

SG

COM

BNP

ISPBCP

BES

BOI

UCR

BES

ISP

AIB

BCP

BBVA

BS

ABN

AIB

BBVA

ING

BOI

BS

UCR

COM

DB

BNP

SG

Before: at day 1 no significant effect of a shock on the sovereign CDS spreads in the case of BNP, SG, ING, BBVA, and BS. At day 22all are significant. After: at day 1 no significant effect of BOI on IR. In the long-run (day 22) no significant effect on the sovereign for ashock from BNP, SG, COM, DB, AIB, BOI, UCR, ABN, and BS. Additionally, there is no effect of a shock from SG on itself at day 22.

The second result of our paper is that the importance of sovereign shocks augmented dramati-cally in the post interventions era. This result is not only backed by generalized impulse responses(cf. Figure 3, especially left panel) but also by the long-run and short-run analysis22. By compar-ing the results from the GIR analysis, we are able to distinguish between financial institutions thatare more sensitive to a shock of their government CDS spread. Figure 4 depicts the comparisonamong selected banks.

Figure 3: Impulse from the Government Sector: Effects on Sovereign CDS (y-axis) and Bank’s CDS (x-axis):at day 1 (Left) and day 22 (Right).

.6.8

11.

21.

4S

tate

− R

espo

nse

(Im

puls

e V

aria

ble)

0 .2 .4 .6 .8 1Bank − Response

Before After

BOI

AIB

ISP

BSBBVA

UCRBES

BCP

BNPSG

DBING

COM

ABN

UCR

BCP

ISP

BES

COMDB

ING

AIB

BNP

ABN

SG

BOIBS

BBVA 0.5

11.

5S

tate

− R

espo

nse

(Im

puls

e V

aria

ble)

0 .5 1 1.5Bank − Response

Before After

ISP

BOIAIB

BCP

BES

BBVA

BS

UCR

ING

BNPDB

SG

ABN

COM

ISP

UCR

BCP

BES

COM

BSBBVA

BOI

BNPAIB

SG

ING

DBABN

Before: at day 1 there is no significant impact of the sovereign shock for ISP, ABN, and BBVA. At day 22 there is no significant effecton DB, AIB, BOI, ABN, BBVA, and BS. Furthermore, a shock of DE on itself is not significant as well. After: all shocks have significanteffects in the short- and long-run.

Sorting banks by the effect (in the long-run, e.g. after 22 days) of a sovereign shock we obtainthe following ranking (from lowest effect to the highest): SG, ABN, DB, BNP, COM, ING, BS, UCR,BBVA, BCP, BES, AIB, BOI, ISP. Long-run responses of the banks from the same country are clus-tered. The only exception are the Italian banks, in which case ISP is more sensitive to a sovereign

22The results from cointegration analysis will be presented in the following subsection for three individual countries.

16

Figure 4: Banks Responses to a Sovereign Shock: Stage 4+5+6 (After Government Interventions).

0.5

11.

5

0 5 10 15 20day

BNP COM AIB UCR ABN

SG DB BOI ISP ING

BCP BES BBVA BS Average

Note: The “Average” line represents the mean of the bank responses from a shock in the seven sovereign CDS spreads.

shock than UCR (146% compared to 75%). While SG is impacted only 35%, the top three banks’ im-pacts range from 128% to 146%. ISP and Irish banks respond the most to a sovereign shock. Theyare followed by the Portuguese and then the Spanish banks, where UCR ranks between BBVA andBS. At the bottom of this ranking are the Dutch, the German, and the French banks. Furthermore,we find that banks that received state capital injections (ISP and COM) are more sensitive to thesovereign shocks than the banks (UCR and DB) that were not recapitalized by the state.

4.2 Specific Country Analysis

In this subsection the results for three countries are presented, i.e. Germany, Ireland, and Italy.These have been selected out of the group of countries considered as they differ strongly in thetotal commitment to the financial sector relative to their 2008 GDP. Ireland presents the countrywith the highest and Italy with the lowest engagement. Germany can be argued to range in themiddle of these measures.

4.2.1 Germany

In the case of Germany we analyze the bivariate setups of the German (DE) sovereign CDS spreadin relation with the CDS spread of Commerzbank (COM) and with the CDS spread of DeutscheBank (DB) respectively. The results for the tests on Granger-causality are depicted in Table 2, forcointegration relations in Table 3, and for impulse responses in Table 4 which are presented in theAppendix.23

23Test results and the graph of the German sovereign CDS together with the German banks’ CDS time series arepresented in Appendix B.2.

17

Cointegration and Granger-Causality AnalysisFor the entire period before government interventions (i.e. Stage 1+2) we find evidence for a stablelong-run equilibrium relationship between the German CDS spread and both bank’s CDS series.The hypothesis that both estimated β coefficients for the banks are equal to −1 cannot be rejectedusing a standard t-test. The error correction equation, e.g. for the relation with COM can bewritten as in the following:

cdsDE,t = 1.083(0.170)

× cdsCOM,t − 2.784(0.685)

− ect, 24

where ect refers to the value of the long-run relation at time t and standard errors are providedin parentheses. As variables are measured in logs the β coefficients may be interpreted as elastic-ities, yielding an equation that could be named as a bank-sovereign CDS relation. This relationimplies, of course neglecting the rest of the estimated dynamics in the model, that a 1% increasein the CDS spread of COM leads to a 1% increase in the CDS spread of DE. For DB the sameinterpretation applies.

The loading coefficients indicate that the German CDS spread adjusts much faster (i.e. |α̂DE | =| − 0.05| > |0.029| = |α̂COM |) to deviations from the long run equilibrium than the CDS spreadof COM. The α-coefficients in the relation with DB suggest that the latter CDS spread does notadjust to any deviations from the long-run equilibrium, while the German CDS spread is adjust-ing with a rate of α̂DE = −0.122. A formal test confirms this result as cdsDB,t is tested to beweakly exogenous, which leads to the argument that DB provides the stochastic trend in this coin-tegration relation. Tests for Granger-causality indicate that only COM Granger-causes DE on a 1%significance level in the period before state interventions.

After government aid schemes are set up the long-run relations change. Firstly we do not finda stable long-run relation for DE-COM for the entire post-intervention period, but only in Stage 5.Compared with the pre-intervention results, we find equal values for the β-coefficients, implyingthe same elasticities as mentioned above. However the constant changes from 2.28 (before) toinsignificant (after) yielding the interpretation that the gap between the two CDS series vanished.In contrast we do find a cointegration relation for DE and DB for the entire post-interventionperiod. This relation implies, again not taking into account the entire dynamics of the model, thata 1% increase in CDSDE,t causes an about 3.4% increase in CDSDB,t. Or put differently a 1%increase of the DB CDS series leads to a 0.29% rise of the German CDS spread, implying that theDB (DE) CDS series has become more (less) sensitive to changes in the DE (DB) CDS spread.

The relation for COM and DE in Stage 5 yields to the conclusion that COM is weakly exoge-nous and DE adjusts with a rate of α̂DE = −0.045, which is close to the α-value from the periodbefore interventions. In the second cointegration relation, that takes together all three stages afterthe bailout scheme, we find DE to move the equilibrium in the direction of its development (asDE’s α and β coefficients are both positive). Granger-causality tests indicate further that in theperiod during and after the state interventions that all variables Granger-cause each other on the

24The cointegration graph is provided in Appendix B.2, Figure 10.

18

1% level.

Impulse Response Analysis

Figure 5: Generalized Impulse Responses for Germany: Before (solid) and During & After GovernmentInterventions (bold)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0 2 4 6 8 10 12 14 16 18 20 22

-0.8

-0.4

0.0

0.4

0.8

1.2

1.6

0 2 4 6 8 10 12 14 16 18 20 22

.0

.1

.2

.3

.4

.5

.6

.7

.8

.9

0 2 4 6 8 10 12 14 16 18 20 22

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

0 2 4 6 8 10 12 14 16 18 20 22

DE -> DE

DE -> COM COM -> COM

COM -> DE

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0 2 4 6 8 10 12 14 16 18 20 22

-0.8

-0.4

0.0

0.4

0.8

1.2

1.6

0 2 4 6 8 10 12 14 16 18 20 22

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

0 2 4 6 8 10 12 14 16 18 20 22

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0 2 4 6 8 10 12 14 16 18 20 22

DE -> DE

DE -> DB DB -> DB

DB -> DE

Note: [Left Panel] Upper-Left: DE (impulse variable) - DE (response variable). Lower-Left: DE (impulse var.) - COM (response var.).Upper-Right: COM (impulse var.) - DE (response var.). Lower-Right: COM (impulse var.) - COM (response var.). [Right Panel]Upper-Left: DE (impulse var.) - DE (response var.). Lower-Left: DE (impulse var.) - DB (response var.). Upper-Right: DB (impulsevar.) - DE (response var.). Lower-Right: DB (impulse var.) - DB (response var.). Solid lines: responses before government interventions(bold) and the 95% bootstrapped confidence interval (thin). Dotted lines: responses during and after government interventions (bold)and the 95% bootstrapped confidence interval (thin). X-axis: number of days (after the shock). Y-axis: impact relative to one standarddeviation shock of the impulse variable.

The results from the impulse response are depicted in Figure 5. The analysis before governmentinterventions refers to the VECM setup; only in the period after we use a VAR framework forexamining the relationship between DE and COM. In all graphs the three solid lines representthe impulse responses before interventions, where the light ones refer to the 95% bootsrappedconfidence interval. The bold dotted line describes the responses during and after rescue schemesand the light dotted lines the bootstrapped confidence bands. Firstly, we observe that the patternof the left panel resembles strongly the pattern depicted in the right panel. In the upper-rightcorner of each panel the effects of a shock from the bank CDS spread to German CDS are plotted.Before interventions (solid) a banking sector shock impacts permanently the government CDSseries, while in the period after (dotted) there is only a temporary effect.

For the case of a government shock to the banking sector we notice that DB (right panel) is onlyaffected in the very short-run (t ≤ 3) before interventions, while COM is driven permanently bythe latter shock. In the period after the state bailouts, we find that both series react permanentlyto a shock stemming from the sovereign.

Additionally, the graphs show that the effects of a banking sector shock on itself are stronger inthe pre-interventions period, as they are estimated to have a permanent effect. The responses after

19

state interventions suggest a decrease of the impact of the latter in both cases. The shocks fromthe government CDS spread on itself have a stronger contemporaneous impact in both bivariatesetups after interventions.

DiscussionFrom October 2008 until the end of May 2010, Germany provided a total support to the localfinancial sector of EUR 619.1bn or of 25% of total 2008 GDP. From a total committed amount ofEUR 64.8bn for capital injections, EUR 54.5bn were demanded by German banks until the end ofMay 2010. Germany pledged EUR 475bn in form of liability guarantees, from which local banksutilized EUR 185.8bn until the end of our time frame.

SoFFin25 granted COM26 an individual guarantee for issuing EUR 15bn of debt securities27.Although DB, the biggest German bank, resisted to state capital injections, we do not find strongdifferences in the dynamics of both CDS series in relation with the German sovereign CDS spread.Nonetheless, our results suggest that investors anticipated the direct support measure for COMas Granger-causality tests underpin that the CDS spreads of COM contain important informationfor determining the price of the German government CDS. Furthermore, a shock to the sovereignCDS drives the CDS spread of COM permanently. Thus before interventions we have evidencethat the dynamics of the series differ, suggesting that the the link between the CDS series of COMand DE is more sensitive than the link between DB and DE.

The research design of our analysis enables us to draw only inference on outcomes. COM isknown to have had severe difficulties during the last crisis, which led, as stated above, SoFFin toprovide extra measures to this bank. The results of our empirical analysis, however, underlinethat the dynamics of the two banks do not substantially differ in the post-intervention period.Assuming that this similarity is a consequence of the extra measure provided, we conclude that therescue schemes in the case of Germany were successful. The extra funding for COM was necessaryin order to induce a credible perception that the tail risk of the latter was absorbed by the state.We find that shocks of both banks have a weaker effect on themselves after bailout schemes areput into place. However, the result is stronger for DB. The cost for this positive aspect is a highersensitivity of both banks to developments in the government CDS spreads. Interestingly, the stateCDS spreads are not influenced at a long horizon by a banking sector shock after bailout measuresare provided.

Altogether, the results highlight that the contagion emerged from the banking sector andspilled-over to German sovereign CDS spreads in the period before rescue schemes. Thus, wefind evidence for H1. The dependence in the other direction is weaker or only existent in the veryshort-run. Afterwards, future developments of the perceived default risk of all series are strongly

25The German Special Fund for Financial Market Stabilization (SoFFin) is in charge for managing the German finan-cial support programs.

26https://www.commerzbank.de/en/hauptnavigation/aktionaere/service/archive/ir-nachrichten_1/2008_5/ir_nachrichten_detail_08_2203.html

27Furthermore, SoFFin participated with EUR 8.2bn in form of a silent equity holding and COM’s recapitalizationamounted EUR 10bn. These capital injections became public on 3 November 2009.

20

https://www.commerzbank.de/en/hauptnavigation/aktionaere/service/archive/ir-nachrichten_1/2008_5/ir_nachrichten_detail_08_2203.htmlhttps://www.commerzbank.de/en/hauptnavigation/aktionaere/service/archive/ir-nachrichten_1/2008_5/ir_nachrichten_detail_08_2203.html

interwoven as suggested by cointegration analysis and the results of the Granger-causality tests.Furthermore, impulse responses highlight a stronger interdependency of all series, while an unex-pected change in the bank’s CDS series has only a temporary effect on the sovereign CDS spread(H2a, H2b). Moreover, we find differences in the dynamics of COM and DB in relation with theGerman government CDS series (H3). However, our results suggest that the extra measures pro-vided to COM have credibly transfered the tail risk on the government’s balance sheet.

4.2.2 Ireland

Within the set of analyzed countries, the results for Ireland reveal most clearly the impact of gov-ernment interventions. As the dynamics for both setups, i.e. Ireland (IR) - Allied Irish Banks (AIB)and Ireland (IR) - Bank of Ireland (BOI), resemble strongly we report only one of them. Tests onGranger-causality are depicted in Table 2, for cointegration relations in Table 3, and for impulseresponses in Table 4 which are presented in the Appendix.28

Cointegration and Granger-Causality AnalysisThe cointegration analysis uncovers a long-run relation in Stage 2, in which β̂AIB = −0.567. Inter-preting the cointegration coefficients for the period before interventions, a 1% increase in the banksCDS translates into a 1.76%29 gain in the Irish CDS. The gap (the constant from the cointegrationequation) between the two cds series is estimated to be insignificant.

Furthermore, in the period before government interventions there is evidence that the stochas-tic trend originates from the banking sector and impacts the sovereign CDS series. The estimatedα coefficient for AIB is not significantly different from zero and the hypothesis of weak exogeneityfor the banking sector series cannot be rejected. Thus, we conclude that the series of AIB influencesIR in the long-run. In the short-run, tests for Granger-causality are not reported to be significantin both directions.

During and after interventions the dynamics change and now emphasize a different role ofthe Irish CDS spread, which we argue occurs because of government interventions. The errorcorrection equation can be written as

cdsIR,t = 0.724(0.105)

× cdsAIB,t + 1.116(0.587)

− ect.30

Comparing elasticities, we now find an increase of β̂AIB to 0.724, implying that a 1% increasein the Irish CDS spread leads to a change of the bank’s CDS spread by 0.724%. The gap betweenthe two series is enlarged and it is significantly different from zero.

The estimated α-coefficients suggest that during and after interventions the Irish CDS spreadprovides the stochastic trend, as a test for weak exogeneity of this series cannot be rejected.Only the bank’s CDS spread adjusts to deviations from the long-run equilibrium with a rate of

28Preliminary test results and the graph of the respective time series are presented in Appendix B.3.29This number is obtained by normalizing the coefficients by the estimated β-coefficient of AIB.30The cointegration graph is provided in Appendix B.3, Figure 12.

21

α̂AIB = 0.06. The prominent role of the Irish CDS series is also emphasized in the short-run dy-namics, where we find that the CDS spread of IR Granger-causes the CDS spread of AIB but notvice versa during this period.

Impulse Response AnalysisThe generalized impulse response, which are depicted in Figure 6, underline the shift in the de-pendency between the two CDS series. Firstly, the graph in the upper right corner indicates thata shock from the banking sector permanently influences the government CDS spread before in-terventions, and only temporarily (t ≤ 2) in the second period. The opposite pattern is foundfor a government sector shock. In the pre-intervention period the graph in the lower left cornerhighlights that the latter shock only temporarily influences the CDS spread of AIB, while there isa permanent impact in the period during and after the rescue schemes. Moreover, the remainingtwo graphs (upper left and lower right corner) suggest that there has been a change in sensitivityto a shock from the same sector. A banking shock yields a permanent effect on itself before inter-ventions and is estimated to incur a strongly decreasing impact. For the Irish CDS spread the GIRresults show an opposite development. Whilst both deviations are permanent, the one during thesecond period is by far stronger.

Figure 6: Generalized Impulse Responses for Ireland: Before (solid) and After Government Interventions(bold)

0.0

0.4

0.8

1.2

1.6

2.0

2 4 6 8 10 12 14 16 18 20 22

-0.8

-0.4

0.0

0.4

0.8

1.2

2 4 6 8 10 12 14 16 18 20 22

-0.4

0.0

0.4

0.8

1.2

1.6

2.0

2 4 6 8 10 12 14 16 18 20 22

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

2 4 6 8 10 12 14 16 18 20 22

AIB -> AIB

AIB -> IR

IR -> AIB

IR -> IR

Note: Upper-Left: IR (impulse variable) - IR (response variable). Lower-Left: IR (impulse var.) - AIB (response var.). Upper-Right: AIB(impulse var.) - IR (response var.). Lower-Right: AIB (impulse var.) - AIB (response var.). Solid lines: responses before governmentinterventions (bold) and the 95% bootstrapped confidence interval (thin). Dotted lines: responses after government interventions(bold) and the 95% bootstrapped confidence interval (thin). X-axis: number of days (after the shock). Y-axis: impact relative to onestandard deviation shock of the impulse variable. Generalized impulse responses for BOI behave similarly to those of AIB.

DiscussionNot surprisingly the study of the Irish risk transfer mechanism depicts most clearly the changein the dynamics, as Ireland (IR) represents, by far, the one with the highest total commitment tothe financial sector relative to its GDP. Remarkably it amounts to 319% of 2008’s GDP; or put in

22

monetary value EUR 592bn. Up to end of May 2010, EUR 99.8bn were required in total by Irishbanks. This amount includes EUR 19.3bn that were used as capital injections (cf. Table 1). Bothbanks in our study were recapitalized by the Irish government on the 21st of December 2008 andapproved by the European Commission on 26/03/2009 (BOI) and 12/05/2009 (AIB).31. Underthis scheme AIB and BOI were each provided EUR 3.5bn. Similar government aid structures forboth banks lead to homogeneous findings, which supports our H3.

State interventions are shown to be most successful in the case of Ireland. The magnitude ofthe rescue scheme has been the highest (relative to GDP) among the countries analyzed, whichled to the clearest results for the risk-transfer mechanism. The impact of a banking sector shockon itself decreases substantially after measures are provided. Furthermore, there is a significantimpact on the government spreads only in the short-run. The flip side of the coin is the stronginfluence of a government sector shock on the banks after the rescue schemes, which amplifiedthe serious issues of the Irish financial sector as the sovereign debt problems emerged.

Combining the results from the two analyses, we find strong evidence for H1, H2a, H2b. Beforebailout programs the data shows that the channel through which risk is spread into the marketoriginates from the banking sector rather than from the government. In the period after govern-ment interventions, the risk transfer mechanism puts more weight on the developments of thegovernment CDS spread. As government took over the tail risk from the banks, the developmentof the Irish CDS series plays an increasingly important role. Only in the very short-run changesin banks’ CDS spreads are found to impact on the government series during and after the stateinterventions. The effects of the a banking sector shock on itself have weakened, underpinningthe success of the Irish bailout schemes.

4.2.3 Italy

The main cointegration relations between Italy and the selected domestic banks (Intesa San Paolo(ISP) and Unicredito (UCR)) are presented in Table 3. Table 2 presents the findings for Granger-causality tests and Table 4 for generalized impulse responses.32.

Cointegration and Granger-Causality AnalysisIn the period preceding the government support for the Italian banking industry (i.e. Stage 1+2),we find that the banks’ and sovereign CDS series are tied together in a long-run equilibrium.Interpreting the β coefficients, neglecting the remaining dynamics of the system, we argue thatin the long-run a 1% increase in a bank’s CDS spread leads to a 0.71% (ISP) or 0.67% increase inthe CDS series of Ireland. The gaps (i.e. the constants in the cointegration relations) between thetwo CDS series is in both setups estimated to be significantly different from zero. The speed ofadjustment, reflected by the estimated α-coefficients, is faster for the CDS spreads of banks, i.e.|α̂IT | = 0.012 < 0.020 = |α̂ISP | and |α̂IT | = 0.010 < 0.014 = |α̂UCR|. Regarding the short-rundynamics results reveal that Italy is Granger-caused by the developments in ISP’s and UCR’s CDS

31cf. IP/09/744 and IP/09/48332Preliminary test results and the graph of the respective time series are presented in Appendix B.4.

23

spread in Stage 1+2, consistent with our assumption that the information from the financial sectorwas systemically important.

After the first Italian government aid program for the financial sector the dynamics betweenthe sovereign and banks’ CDS spreads change. Firstly, UCR is found to be in a stable long-runequilibrium with the Italian government CDS series only during Stage 5. In this setup the esti-mated β-coefficients imply that a 1% increase in the spread of the government incurs an upwardadjustment of UCR’s spread of 1.28%. The error correction mechanism of IT-ISP for the entirepost-intervention period is as follows:

cdsIT,t = 0.864(0.087)

× cdsISP,t + 0.922(0.385)

− ect.33

A marginal change of the Italian CDS series by 1% leads to an adjustment of cdsISP by 0.86%.Elasticities of both banks cannot be compared as they refer to different stages of our sample. Lastly,gaps are significantly different from zero in both setups.

The loading coefficients indicate that Italy provides the stochastic trend in the period aftergovernment interventions, as the latter CDS series is tested to be weakly exogenous. This resultimplies that while the Italian CDS spread does not adjust to deviations from the long-run equilib-rium the banks’ CDS spreads react to these changes. In comparison with the results of the previousperiod, in the period after state interventions the Italian CDS spread Granger-causes both bank’sCDS spreads but not vice versa.

Impulse Response AnalysisThe graph in the upper right corner of each panel depicts the effect of a banking shock to thesovereign CDS series. The solid line emphasizes that risk permanently spreads to the governmentCDS series before interventions, while after interventions the shock of UCR (right panel) shiftsthe Italian CDS series stronger but only temporarily (t ≤ 18). A shock originating from ISP (leftpanel) is stronger, as found for UCR, but incurs a permanent shift of the government CDS spread.These findings support our H2a and H3. On the other hand, the GIR analysis of a shock fromthe government sector (the lower-left graph in each panel) reveals that after a movement in thevery-short run the effect becomes insignificant in the period before state interventions (solid lines),but then become permanent after t = 5 (UCR) and t = 8 (ISP). After interventions (dotted lines)the impact is stronger and permanent, in line with our H2b. A shock on the series themselveshave a similar impact on themselves (the lower-right corner in each panel). For the case of abanking sector shock on itself, the results even suggest that a shock before has been about thesame importance. A government shock on itself is stronger in the period afterwards for bothsetups, but temporary in the case of UCR.

DiscussionItaly has one of the highest debt burdens34 among European Union countries. This fact determinedthe Italian government to pledge in total EUR 62bn, that represents slightly above 4% of the 2008

33The cointegration graph is provided in Appendix B.4, Figure 1434Italy’s public debt was estimated around 105% of GDP in 2008.

24

Figure 7: Generalized Impulse Responses for Italy: Before (solid) and After Government Interventions(bold)

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

0 2 4 6 8 10 12 14 16 18 20 22

0.0

0.2

0.4

0.6

0.8

1.0

1.2

0 2 4 6 8 10 12 14 16 18 20 22

-0.4

0.0

0.4

0.8

1.2

1.6

2.0

0 2 4 6 8 10 12 14 16 18 20 22

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0 2 4 6 8 10 12 14 16 18 20 22

ISP -> ISPIT -> ISP

IT -> IT ISP -> IT

0.6

0.8

1.0

1.2

1.4

1.6

1.8

0 2 4 6 8 10 12 14 16 18 20 22

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

0 2 4 6 8 10 12 14 16 18 20 22

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0 2 4 6 8 10 12 14 16 18 20 22

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0 2 4 6 8 10 12 14 16 18 20 22

UCR -> UCRIT -> UCR

IT -> IT UCR -> IT

Note: [Left Panel] Upper-Left: IT (impulse variable) - IT (response variable). Lower-Left: IT (impulse var.) - ISP (response var.).Upper-Right: ISP (impulse var.) - IT (response var.). Lower-Right: ISP (impulse var.) - ISP (response var.). [Right Panel] Upper-Left: IT (impulse var.) - IT (response var.). Lower-Left: IT (impulse var.) - UCR (response var.). Upper-Right: UCR (impulse var.) -IT (response var.). Lower-Right: UCR (impulse var.) - UCR (response var.). Solid lines: responses before government interventions(bold) and the 95% bootstrapped confidence interval (thin). Dotted lines: responses after government interventions (bold) and the 95%bootstrapped confidence interval (thin). X-axis: number of days (after the shock). Y-axis: impact relative to one standard deviationshock of the impulse variable.

GDP. This ratio is the lowest among the analyzed countries in this paper. Capital injections werein sum of EUR 4.1bn from a committed amount of EUR 12 bn. Italy also promised to support itsdomestic banks with an asset purchase scheme worth EUR 50bn. On 20 March 2009 ISP receivedEUR 4bn as state aid for recapitalization. ISP issued subordinated debt that was subscribed bythe Italian Treasury. On the other hand, UCR which is the biggest Italian bank did not receiveany capital injection from the state of Italy. The different treatment of the banks is reflected in theGIR analysis, however yielding support, that the Italian CDS series became permanently moresensitive to the ISP spread. This gives evidence for our third hypothesis (H3).

Thus, in the case of ISP, we find evidence that the rescue measures taken by the Italian gov-ernment were not successful in absorbing partial risks of default from the latter bank. A shock ofthe bank on itself has even a stronger effect after bailout schemes. The other finding underpins,that investors believed that default risk from ISP would spread to the government sector. Only forUCR, we find a similar pattern as in other countries, even though we detect only a small decreaseof the effect of a banking sector shock on itself.

Before Lehman Brothers’ default, the systemic banking crisis spreads to the sovereign market,which can be seen by the results from Granger-causality analysis, or the permanent effect of abanking sector shock in the context of the GIR analysis. However, movements of IT’s CDS spreadhave an effect on the bank spreads as well, which contradicts partially our H1. After state inter-ventions, this relation becomes more pronounced as now IT Granger-causes both banks, provides

25

the stochastic trend in the cointegration relations, and government shocks cause strong deviationsin the banks’ CDS series. Nonetheless, banks still influence the government CDS series, whileUCR only temporarily. Government packages seem not to have limited the effects of a bankingsector shock on itself as the intensity of the impact is almost the same as in the period before gov-ernment interventions. On the other hand, the sovereign CDS series turn out to be more sensitiveto shocks as a result of government support schemes.

5 Conclusions

The recent financial crisis led governments to tailor aid programs for financial institutions to steeragainst the failure of the entire system. The magnitude and dimension were unique in Europeanhistory. A series of bank failures would threaten the functioning of the whole economy, as theimportant financial institutions incorporate a systemic component. Hence, governments, besidescentral banks, took crucial steps in the attempt to rescue the financial system. This paper proposesone way to underline the changes in interdependencies of banks and sovereign credit spreadsin the euro area. By arguing that government bailout programs marked an important event forinvestors, we derive our hypotheses that reflect how the relations are expected to change. First ofall, we hypothesize that the increase in default risk prior to interventions originates mainly fromthe financial sector. After bailout programs are set up by European governments, we argue that thesensitivity of the sovereign default risk to the financial sector increases due to the private-to-publicrisk transfer. Moreover, the default risk of the banking sector is asserted to be influenced stronglyby the government sector. How deeply a bank participates in the rescue scheme should affect itsCDS sensitivity to changes in the sovereign spread. Finally, we argue that important determinantsfor the changes in linkages are country specific characteristics related to bailout programs. Weconduct our empirical analysis by using data of seven European sovereign CDS series togetherwith two local banks’ CDS series for each country. We analyze CDS spreads because we considerthem as a proxy for default risk and divide the sample into before and after bailout programs. Ifstructural breaks are detected we use sub-periods based on common economic trends related tothe financial crisis. In order to find support or counter-evidence for our formal theories, the studyemploys techniques from cointegration, Granger-causality, and impulse responses analysis.

As stated in our first hypothesis, before government interventions, sovereign credit risk isstrongly impacted by the movements in bank CDS spreads, while changes in the sovereign CDSspreads have a weak impact on both bank and state CDS markets. Our findings provide evidencefor H1 for FR, DE, IR, NL, and SP but it is contradicted by our results for IT and PT. Portugal’s andItaly’s default risk seem to carry an important role in the development of their local banks’ defaultrisk even before the Lehman Brothers event.

For the second set of hypotheses (H2a, H2b), i.e. after government interventions, we can con-clude homogeneously that changes in the sovereign CDS spreads contribute to the financial sectorrisk of default developments. On the other hand, the changes in banks’ risk of default are found to

26

impact stronger in the short-run (i.e. at day 0 and at day 1) in all countries, while in the long-run(i.e. after 22 days) the effects are transitory (exception IT, NL, and PT).

Furthermore, the variability in interdependencies between bank and sovereign CDS spreadscan be related to differences in state support measures accessed by each bank. Countries (FR,IR, SP, and PT) with similar state aid for both analyzed banks show an equal sensitivity to thesovereign CDS spread. Moreover, as banks in Germany (DB and COM) and Italy (ISP and UCR)were differently involved in the rescue schemes, we also find heterogeneity in the linkages be-tween bank and sovereign CDS spreads. Our results suggest that the extra aid provided to COMhas been successful in absorbing partially the risk of default while the extra measures for ISP linkthe default risk of the latter strongly to the development of the Italian CDS spread and amplifyits sensitivity to shocks in both the banking and the sovereign sector. Furthermore, our resultsindicate that bailout schemes in the case of Ireland have led to the desired results.

Lastly, the cross-country analysis reveals the heterogeneity of the impact of government sup-port programs. On the one hand, the effects of a sovereign shock to banks from the same countryare closely linked, and on the other hand the effects of a sovereign shock to banks across countriescan be clustered in INNER (FR, DE, NL) and OUTER (IR, IT, PT, SP).

27

References

ATTINASI, M. G., C. D. CHECHERITA AND C. NICKEL, “What Explains the Surge in Euro AreaSovereign Spreads During the Financial Crisis of 2007-09?,” ECB Working Paper (2009).

BERNDT, A. AND I. OBREJA, “Decomposing European CDS Returns,” Review of Finance 14 (2010),189–233.

BIS, 79th Annual Report (Basel: Bank for International Settlement, 2009).

———, Quarterly Review (Basel: Bank for International Settlement, December 2008).

CASTREN, O. AND I. K. KAVONIUS, “Balance Sheet Interlinkages and Macro-Financial Risk Anal-ysis in the Euro Area,” ECB Working Paper (2009).

COLLIN-DUFRESNE, P., R. GOLDSTEIN AND J. MARTIN, “The Determinants of Credit SpreadChanges,” Journal of Finance 56 (2001), 2177 – 2207.

DIECKMANN, S. AND T. PLANK, “Default Risk of Advanced Economies: An Empirical Analysisof Credit Default Swaps during the Financial Crisis,” Working Paper (2010).

DOLADO, J. J. AND H. LÜTKEPOHL, “Making Wald Tests Work for Cointegrated VAR Systems,”Econometric Reviews 15 (1996), 396–386.

DÖTZ, N., “Time-Varying Contributions by the Corporate Bond and CDS Markets to Credit RiskPrice Discovery,” Deutsche Bundesbank Discussion Paper (2007).

DÖTZ, N. AND C. FISCHER, “What Can EMU Countries’ Sovereign Bond Spreads Tell Us AboutMarket Perceptions of Default Probabilities During the Recent Financial Crises?,” Deutsche Bun-desbank Discussion Paper (2010).

EJSING, J. AND W. LEMKE, “The Janus-Headed Salvation: Sovereign and Bank Risk Premia during2008-2009,” Working Paper (2009).

ENGLE, R. AND C. GRANGER, “Co-Integration and Error Correction: Representation, Estimation,and Testing,” Econometrica 55 (1987), 251 – 276.

FONTANA, A. AND M. SCHEICHE, “An Analysis of Euro Area Sovereign CDS and Their Relationwith Government Bond,” ECB Working Paper (2010).

FORTE, S. AND J. I. PENA, “Credit Spreads: An Empirical Analysis on the Informational Contentof Stocks, Bonds, and CDS,” Jou